Summary: This post introduces Bezca, a CALL software prototype for training Chinese characters. It showcases that (1) the technology in Finnish Annotator can be modified to many different purposes, (2) having a day job does not prevent me from writing CALL software with realistic goals, and (3) context is fairly easy to integrate to any CALL software, as long as you take an uncompromising attitude towards the need for context.
In Bezca, you train Chinese characters by drawing the strokes with a stylus, finger of plectrum. Correctly drawn strokes appear to the screen as you draw them:
If you don't remember the character, you can look at the hint which also tells the next stroke:
Clicking "Show Examples" displays dictionary words and examples sentences which use the character in question.
In the example browser, if you don't know some word in an example sentence, you can just click on it and enjoy further examples about that word. This way, you can browse examples in an endless chain, Wikipedia style.
Bezca also contains a spaced repetition system. Pressing "This was easy" shows the exercise again in a week. Pressing "This was hard" shows it again in a day. After that, it uses exponential time gaps to decrease or increase the period based on user responses.
In the beginning, the program calibrates difficulty to suit the student's skill level. It shows some characters and asks the user to say if they are suitably difficult or too easy. This way, students can go directly to material which is new for them.
This is a prototype and not yet mature enough to be distributed. It contains only 180 characters. Installing requires a memory card. I don't currently have any plans to take it further, because demand is likely to be small, the use of CEDICT would make it a copyright violation to ask for a price, and it would take a lot of effort to input 1000 characters. However, I'm happy to demonstrate it face-to-face to anyone, especially CALL researhcers.
Summary: This very dry post summarizes which areas of Chinese learning are adequately covered by CALL tools, and which areas still need better tools and content.
Hamming's essay for choosing research topics describes CALL spot on, if you replace "research paper read by thousands" with "learning system used by thousands". Finnish Annotator would have passed Hamming's scrutiny, since annotators had already proven useful in Japanese and Chinese, but none was available for Finnish. It aimed at the core of Finnish reading comprehension.
The downfall of FA was partly due to inadequate openness, feedback and networking. This was also predicted in Hamming's essay:
Some people work with their doors open in clear view of those who pass by, while others carefully protect themselves from interruptions. Those with the door open get less work done each day, but those with their door closed tend not know what to work on, nor are they apt to hear the clues to the missing piece to one of their "list" problems. I cannot prove that the open door produces the open mind, or the other way around. I only can observe the correlation. I suspect that each reinforces the other, that an open door will more likely lead you and important problems than will a closed door.
Let's take Hamming's advice to the conclusion and make a list of important problems in computer-aided language learning. This list only covers Chinese, which has the special challenge of learning the characters. It also ignores collaborative learning methods and concentrates on single-user teaching machines.
Why ignore collaborative learning?
The currently dominant learning theory is Communicative Language Teaching (CLT). People use language to achieve communication goals like buying a ticket or describing a problem. CLT claims that also in teaching, each sentence should be part of a speech act with a communicative aim. Modern first-year language textbooks achieve communicative context by describing situations, where tourists achieve communication goals.
CLT is trivially true in the sense that sooner or later you have to move from isolated sentences to communication, for example talking, email exchange or searching for information (and not just reading for the sake of language). However, you have a long way to go before you can read books or write blogs. Before CLT forces itself through the door, you have to bootstrap the language skill somehow. I'm not at all convinced that CLT is necessary in the initial phase. The situation I see on the ground is that the Japanese/Chinese self-study scene is blithely unaware of CLT and still achieves good results.
Don't get me wrong: communicative context is nice, and the best kind of context you can have. But it is hard work to achieve communicative context. You have to make compromises in other areas. CALL scene is nowhere near the level where the presence or absence of communicative context would make a difference.
Finally, a word of warning if you try to achieve communicative context by collaborative learning. B.F.Skinner, the father of behaviorism described the problems of collaborative learning methods already in 1953. What's the point of making CALL tools at all, if you just digitize the same old problems?
Skinner's children were growing up. When the younger was in fourth grade, on November 11, 1953, Skinner attended her math class for Father's Day. The visit altered his life. As he sat at the back of that typical fourth grade math class, what he saw suddenly hit him with the force of an inspiration. As he put it, "through no fault of her own the teacher was violating almost everything we knew about the learning process." In shaping, you adapt what you ask of an animal to the animal's current performance level. But in the math class, clearly some of the students had no idea of how to solve the problems, while others whipped through the exercise sheet, learning nothing new. In shaping, each best response is immediately reinforced. Skinner had researched delay of reinforcement and knew how it hampered performance. But in the math class, the children did not find out if one problem was correct before doing the next. They had to answer a whole page before getting any feedback, and then probably not until the next day. But how could one teacher with 20 or 30 children possibly shape mathematical behavior in each one? Clearly teachers needed help. That afternoon, Skinner constructed his first teaching machine.
Summary
Area
Status
Method is
Reading, 0 - 1000 characters
Jury is still out on correct approach
Mixed
Reading, 1000 - 3000 characters
Solution is known but not implemented
Spaced repetition systems with immersive sentence decks
Reading, 3000+ characters
Solution implemented, room for improvement
Reading natural texts through an annotator and using example sentence search for new characters and phrases
Writing
Not even started
Translation sentences, chatbots (neither exists)
Listening
Solved
Listening internet radio or simplified podcasts
Speaking
Solved
Talking face to face or through Skype
Beginner phase: 0 - 1000 characters
First of all, beginners and advanced students should use very different methods. When advanced students learn a phrase, it integrates naturally with their existing knowledge. They can immediately use the word in different contexts. Beginners are only forming those knowledge structures.
For valid historical reasons, current CALL tools are not very good for beginners. In many cities elementary courses are available for Chinese and Japanese, but courses stop after that. Beginner phase also lasts for shorter period. Therefore there is less demand and less tool development for elementary tools. In the intermediate and advanced phase, it is important that the tools scale and can teach large amounts of phrases and accommodate different skill levels. This also means that a software package only needs to implement one scalable method well, for example dictionary search or flashcards.
For beginners, my unjustified gut instincts is that learning games like Slime Forest Adventure are the way to go. (1) Beginners forget things more quickly, since their knowledge structures are just forming. Therefore intensive teaching methods are good and immersive approaches which give little time to forget are preferable. (2) Beginners need to look at the language from several different perspectives (sentence comprehension, syntax, word inflection, communication) all of which are completely new to them. Game programming has the tradition of subgames, which have their own set of rules. I don't see such tradition of variability in other types of software.
Reading comprension in 1000 - 3000 characters
The software is there, but content has plenty of room for improvement. Annoatotrs enable reading easy texts and spaced repetition systems with sentence decks are good for learning characters. Regarding content, I haven't seen any easy reader texts except in Chinesepod. The sentences in my HSK deck were pretty random: they were ripped from the example sentence collection in an online dictionary and then automatically classified by difficulty.
3000+ characters
At this point you can read natural texts and start to read for content. An annotator and example sentence search are all you need. They already exist.
Writing
I haven't met any CALL tools for training writing skill. The only method is to "jump to the water and swim" by just starting to write emails and blog posts. This is comparable to practising reading comprehension by just taking a dictionary and a foreign-languge book. Sure, you can do that, but it requires a lot of motivation and willpower.
Listening
There are many free radio stations available, and Chinesepod offers easier dialogs. You can listen to them while you clean or cook. There is nothing to improve, since we are already at zero time commitment. This is the ultimate in efficiency.
Speaking
Speaking is the only way to learn to speak. I don't see how CALL tools could play any role in this. Skype already works.
In Tampere University Alakuppila cafe ther are regular meetings, where Chinese exhange students talk with Finnish language students. For those who live in less forutnate places, there are various commercial services, some of which offer free samples.
Finnish Annotator was my CALL website, developed around 2005-2006. In those years, I was finishing my studies and spent summers writing the website. The site featured an annotator for Finnish and Chinese, a flashcard program and a character-drawing exercise. I took it down in 2008 as it had no users.
Annotator is a "text dictionary", which decodes the inflection and searches explanations for all words in a copy-pasted text. While Google Translate is free, annotators are more useful for language-learners. You can read the text as long as you completely understand it, resorting to hovering your mouse over annotations only when you have to.
The entry page shows how it annotated Chinese text. It also describes how you could turn a copy-pasted text into a flashcard deck. The post about the fundamental problem of flashcards mentioned that my website tried to solve it by taking example sentences from the annotated text. Indeed after you you press "show answer" it showed annotated example sentence where "kun" was used.
The Finnish vocabulary contained 1000-word test vocabulary. The demo page used to work on all browsers, but currently crashes Firefox. Being acutely aware of the need for context, the word definitions contain well-split meanings and example phrases, and sometimes even comparison and contrast to related words.
At the bottom of the Chinese entry page there is a screenshot of the character-drawing exercise, where you move the brush with your mouse and the stroke appears if you are moving the brush correctly. This mayseem similar to Skritter, since both programs took influence from WriteChinese, a piece of prior art from the nineties.
Morphology engine and master's thesis
About half of the code in the website deinflects Finnish words. Finnish inflection is very complex: for example substantives can have 4 different types of postfixes. The site used two-level morphology and state machines to decode the words. These were a bit obsolete methods to handle morphology, but they were provably successful for Finnish and clearly described in Koskenniemi's book. Modern methods would have required access to commercial state machine libraries, which I didn't have.
My thesis described the algorithms in the morphology engine. It used athematical notations and also contained a few proofs. When I returned it, it got full points.
The algorithm for compiling two-level inflection rules contained a minor simplification. Thesis inspector said that it was actually publishable research, but I didn't follow up on that, since I was not planning to return to school. Anyway, it kind of demonstrates that I already know how to do research, I just don't know how to identify it and wrap it into form, which can be sent to conferences and journals.
How it failed
Since I consider myself economically rational and didn't work for two summers, I had to rationalize away the congnitive dissonance somehow. My feeble excuse was that I was doing a semi-commercial system, which would continue to mill extra income after intial setup effort. In practise, what I did was closer to a mild for of hikikomori.
Firstly, I didn't tell about the system to many people, thinking that I'll publish the product when it is ready. Therefore not a single person becase interested enough in it to give feedback and criticize away obvious weaknesses which were easy to correct but for which I was blind, having spent too much time doing it. For example the need to log in first was such a weakness.
Also, in those days I had not yet discovered the Game of Talking and I kept getting bad outcomes in human relationships without really understanding what the hell went wrong. When I wrote last year "Most people develop these surfacial skills as young adults. Unfortunately, you can't skip the development of social skills. If you fail to complete this developmental task as a young person, it will continue to haunt you and drag you down until you solve it.", I meant also Finnish Annotator. This severly limited my ability to get feedback on the system.
The system was quite close in function not just to MDBG annotator, but also to Lukutulkki, a commercial system for annotating English text to Finnish speakers. Had I presented it right, some CALL researchers should have become interested in it.
The most damaging hit from commercial mindset was my reluctance to use gray copyright vocabularies. It was also a question of quality, as dict vocabularies didn't have split meanings nor example phares. I actually started to collect my own Finnish vocabulary. In the end, it had inflections for about 5000 words and meanings and example phrases for somewhat over 1000 words. At that point, Google Translate published Finnish translation, so I thought that no way in hell am I going to get the vocabulary collected before free services offer better than what I have. Since the system had no users, I took it down. It was really idiotic move to start to collect vocabulary from scratch. I believe now that Takkirauta's talk about Manstein's matrix has a seed of truth, and if you notice that you are doing a lot of repetitive informational work (like vocabulary colllection), you are probably doing something wrong and should stop to ponder different options. Don't just do something, stand there!
The main lessons I learned from it are the importance of social skills and awareness that I am prone to obsessive-compulsive tunnel vision which makes me exert a lot of effort when the right solution would be to look at different options.
What parts of it are still useful
Before I can apply for graduate studies, I need to find a research group. Finnish Annotator is my main merit for persuading others to include me in their work and publications. Next, I'll list examples of how the technologies and components in FA could contribute to CALL research.
The character-drawing engine can be modified to train students to write Russian or Arabic characters. In the first Arabic course I participated, learning to read the script was a huge part of the course. Speeding it up with spaced repetition system, which gradually introduces new material after ensuring that the student has mastered dependencies could make a big enough difference for a publishable paper.
Since the two-level morphology can handle Finnish inflection, it can deal with almost any language. Annotation works best when embedded to other services. FA didn't just annotate copy-pasted text, it also annotated any example sentences in the flashcards. Annotation can be integrated to boost any existing research ambitions in CALL.
ChinesePod is the only commercial language learning service which I have used. In 2009 I subscribed for one year. I was quite satisfied with it. Their service consists of textbook-style lessons. Each lessons is independent and covers one theme. The easiest lessons are targeted at beginners; the hardest one take their text from outside source and assume that the student can read it without aids. They publish several lessons a week.
Textbook chapters are annotated by hand. This way, annotations are correct even when words have several meanings or meaning depends on the context. In addition, there is spoken dialog for each chapter.
In the autumn 2009 I discovered Anki and 20000-word HSK sentence deck, and just stopped using Chinesepod despite having paid subscription. At the time, character recognition was the main obstacle preventing me from reading natural texts, and free tools addressed this problem better. Spaced repetition system was superior to the lessons of Chinesepod.
Great way to make sense of sentence deck sentences and increase reading comprehension after you know enough characters
Companies can put more resources into finalizing their CALL tools. Therefore they have higher quality content. Free CALL tools have two advantages. Firstly, they can use "grey copyright" databases, which are de facto free, although license prohibits commercial use and sometimes also other use.
Secondly, two unrelated individuals can contribute to free tools. Both in Anki and MDBG this plays crucial role. In MDBG, Paul Denisowski initiated the CEDICT vocabulary collection and then disappeared. Someone who prefers to stay anonymous maintains MDBG. Anki was written by Damien Elmers while the 20000-sentence HSK deck was written by Brian Vaughan.
The semi-commercial tool, Slime Forest Adventure, would become better if it was open-source - sooner or later, someone would address the fundamental problem of flashcards and turn it into another great tool. But it possibly wouldn't exist without the profit motive.
Summary: This post tells why Anki and sentence mining are important steps forward in the computer-aided language learning scene. Both steps have happened during the last 3.5 years.
Background: The fundamental problem of flashcard programs
When studying languages, flashcard programs show you a word and ask you to give the translation. In recognition task the program shows the foreign word and asks for the English meaning. Production task tests your ability to spell out the foreign word. Flashcard programs are called spaced repetition systems because they contain timing algorithms which ask easy questions rarely and difficult questions often until they become easy. This ensures that the material is on average suitably difficult.
The fundamental problem is that you don't learn the word by remembering its translation. If you now memorize that Telugu word "adivaramu" means Sunday, you'll just forget it in a few weeks. Spaced repetition systems can delay this to months by reminding you about the word. But to permanently learn a word in the sense that Finnish English speakers know that "Sunday" means "sunnuntai", you need context. You need to see the foreign word in tens or hunderds of sentences, so that it integrates with larger data structures in your head and is no longer just a factlet like "the circumference of earth is 44000km".
This problem is specific to spaced repetition systems, because it is already solved in the analog world. Language textbooks provide the context in the text chapters. Filogists who train to be interpreters and translators mainly read books to expand their vocabulary. In that situation all words are in context.
I first realized this problem after I banged through 1000 Lojban words with Logflash only to forget them all in 3 months.
My first conclusion was that you should only flash cards for which you have text. This worked great with Practical Chinese Reader I & II. First I flashed the words and then I read the text. Thanks to spaced repetition system I could go through chapters much faster.
When I started to build my own language-learning website, I fully realized the importance of tackling this problem. I was using MDBG annotator. It can turn any text into decent study material, unless the text is much above your level. My first approach was to grab the context from the same source as the words. My website had a feature which turned a copy-pasted a Chinese text into a flashcard deck, which contained all words in the text. It also had an easy interface for removing familiar words. The word flascards had context attached: After you gave your answer, it showed the sentences where the word appeared. It also annotated the sentence MDBG-style: When your mouse hovered over any unknown word in the sentences, the meaning of the word appeared.
This solution had a shortcoming: The sentences were too long and difficult, and having just one sentence of context was not enough. I also realized that the real learning happened when studying the sentences, and that they were at least as important as the words being flashed.
My second solution was to collect a database of translated easy sentences and to automatically match them to flashcards. I never properly implemented this, because it required HUGE amount of database collection. Anyone who has ever written example sentences knows how slow it is. The best I achieved was to type enough sentences for an elementary course in Chinese. The material contained Skritter-style character drawing exercises for 200 characters and simple, clear, translated example sentences for them all. This produced adequate quality but it didn't scale. This lack of scalability made it a toy site. Shortly after that, I graduated and stopped developing the site.
Sentence-based flashcards
During the last 3.5 years, an ingenious solution surfaced to the Fundamental Problem: Sentence mining. The idea is that sentences are the basic unit of flashing, not words. Just like gymansts train whole-body movements and just trust that individual muscles get stronger, in sentence flashcards you just trust that you also learn words while flashing sentences.
This is a new developement, as Xamuel's artice is written September 2009 and the Chinese sentence deck I now use was written in 2008. I stopped working on SRS in 2007. This idea is so simple that it makes me ashamed that I didn't notice it. I had already diagnosed the problem and was trying different solutions to it, but somehow failed to take the last step of imagination and to fully move to sentence-based cards.
My own experience confirms that it works like dream. During my Chinese study, I've periodically benchmarked my character count with Clavis Sinica's character test. During the first 4 years, I reached the weekly average score of 2200. During 10 months with sentence deck, the character count exploded to 3000. I could have reached the current skill level a full year earlier, had I known about this method. Now I no longer use the sentence deck, because it has been so efficient that the bottleneck has moved away from single Chinese characters and more context-heavy methods like reading texts with MDBG are more appropriate.
Anki
The rise of Anki is the second big step forward in the computer-aided language learning (CALL) scene. Anki does not contain anything revolutionary, but it combines all good features from all previous flashcard programs into one consitent and easy package. It is so good that if I entered into CALL scene again for the purpose of doing research for graduate studies, I would scrap my old website, which included a spaced repetition system, and use the superior, refined and open-source Anki instead as a basis.
Conclusion
Although my own CALL efforts failed, recent developments in CALL field demonstrate that I was tackling the right questions: How to get context for words in flascards, and how to construct a good spaced repetition system. Progress happened when these problems were addressed. I've witnessed the superiority of the result myself with Anki and 20000-card HSK sentence deck.
That is the title of the first Chinese-language book I've started to read, usually one page a day at bus. It seems to be Harlequin-style romantic book, and that's exactly what I need. Shallow plot is good if you don't understand every sentence. Easy topics like love, sex and success are good for maintaining interest despite slow reading speed.
I chose it, because it was the first book in the shelf for which I recognized all the characters in the title. The second such book was 告诉我你的梦 - Tell Me Your Dreams by Sidney Sheldon. When this Chinese reading binge is over I'll know chick lit really well.
If some girls asks what I'm reading I'll say it's a classic of zen buddhism, since 2/3 of Chinese character tattoos I've seen on girls have included such references.
